YOLOv5 based Open-Source UAV for Human Detection during Search And Rescue (SAR)

Abstract

Floods and earthquakes are among the most frequently occurring natural disasters. They account for high mortality rates due to their rapidness and uncertainty of occurrence. Inundated lands require a quick response for rapid evacuation, arresting fatalities, and consequential economic losses. People tend to seek shelter at dry and open lands at times of calamity. The manual Search and Rescue (SAR) operations have their shortcomings due to the difficulties in identifying the human presence. It requires a longer time for evacuation and therefore increased mortalities. This paper proposes a quadcopter for real-time monitoring of isolated places and automatically detecting stranded humans during floods using image processing techniques at affordable rates. Live video streaming is possible with a camera and a video transmission system attached to the quadcopter. The rescue centers automatically receive the location of humans in case of human detection. Our model integrates an Open-Source autopilot system model, APM 2.8 multicopter flight controller that efficiently stabilizes the flight, and a YOLOv5 object tracking convolutional neural network algorithm for faster detection of human beings. The model is trained using a dedicated dataset of more than 1000 images and attains 0.954 mAP. We have developed a drone using open-source hardware and software tools, conducted test flights to check its stability and the efficiency of the object detection algorithm. We also conducted a mini-survey of one of the most flood-prone areas of Thrissur district in Kerala, using Mission Planner open-source software to evaluate how quickly our drone can assess the entire area. The aim is to save more human lives by quick and efficient aerial assessment in the most cost-efficient manner.